The numerical reactor system for China's new-generation exascale supercomputer

Nuclear energy is a cornerstone for achieving the “dual-carbon” goals and ensuring national energy security, and its sustainable development is predicated on the fundamental premises of absolute safety and economic efficiency. A numerical reactor is essentially the digital twin of a physical nuclear reactor. It leverages next-generation supercomputers to bridge microscopic mechanisms with macroscopic phenomena by performing high-fidelity analysis of the intrinsic coupling of multi-physics fields across multiple scales, thereby establishing a critical digital infrastructure that enables precise, full-lifecycle prediction and virtual experimentation.

The realization of numerical reactors is the in-depth integration of two national strategic pillars: nuclear reactors and supercomputers. For instance, the US El Capitan, currently the world’s fastest supercomputer, is supporting high-fidelity nuclear reactor simulations as one of its primary missions. In recent years, domestic supercomputing systems, represented by Sugon, Tianhe, and Sunway, have entered the exascale computing era. Their powerful computing capability now makes it possible to accurately reproduce the extremely complex phenomena inside reactors. Translating advanced reactor theories across multi-physics and multi-scales into large-scale parallel algorithms that run efficiently on domestic supercomputing platforms and building an independently controllable numerical reactor system have become a strategic cornerstone for ensuring energy security and seizing the global high ground in nuclear technology.

Key challenges in numerical reactors

Developing a numerical reactor capable of accurately reproducing the in-service behavior of a physical reactor is a widely recognized core challenge in the field of nuclear science and engineering. Internationally, advanced nuclear nations have launched large-scale research initiatives. For instance, the US CASL (Consortium for Advanced Simulation of Light Water Reactors) program successfully developed the VERA (Virtual Environment for Reactor Applications) software package for pressurized water reactors while also building the MOOSE (Multi-physics Object-Oriented Simulation Environment) coupling framework under its NEAMS (Nuclear Energy Advanced Modeling and Simulation) program.1 In Europe, with the unified SALOME (Simulation numérique par Architecture Logicielle en Open source et à Méthodologie d’Évolution) platform as the core, research and development outcomes from multiple projects, including the NURE series (such as NURESIM [Nuclear Reactor Simulation]) and McSAFE (High Performance Monte Carlo Methods for Safety Demonstration), have been efficiently integrated.2 However, these cutting-edge explorations collectively point to several key technical barriers that have not yet been fully overcome and that must be solved to achieve the leap from “simulation” to “accurate prediction.”